Tobi Delbruck: Building a high-performance event-based silicon retina
This is a talk given at the Redwood Center for Theoretical Neuroscience, UC Berkeley on February 13, 2007. Speaker Tobi Delbruck, from the Institute of Neuroinformatics, UNI-ETH Zurich.
Run time 74:58Producer Redwood Center for Theoretical NeuroscienceAudio/Visual sound, colorContact Information Kilian Koepsell Redwood Center for Theoretical Neuroscience University of California Helen Wills Neuroscience Institute 132 Barker, MC #3190 Berkeley, CA 94720-3190
Full title: Building a high-performance event-based silicon retina leads to new ways to compute vision.
This talk will describe recent developments in building a high quality spike-based temporal contrast silicon retina vision sensor and using it for vision in real-world situations. The vision sensor will be described, focusing on its emulation of the transient visual pathway in the retina and the way in which it achieves its high performance. Then the talk will turn to novel procedural algorithms for computing low-level vision (e.g. orientation, motion, stereopsis) and high-level object tracking in an event-driven way. It will be argued that the combination of neuromorphic "address-event" representation and digital computation provides a tangible path towards realization of functional neuromorphic systems with distinct advantages compared with traditional approaches.